Mammals first detect microbial infections via an array of PRRs that include both cell surface TLRs and cytosolic NLRs. However, not all microbial interactions represent a pathological state, and the immune system must be able to discriminate to some degree between colonization by commensal organisms and dangerous infection. One level of discrimination is provided by the desensitization or anatomical sequestration of TLRs at sites of chronic stimulation, such as the gut, which presumably allows for tolerance to normal flora 
. Bacterial pathogens can still be recognized at these sites via NLRs, since these systems rely on the specific ability of pathogens to translocate PAMPs into the host cytosol.
The concept that NLRs are specific for pathogenic organisms that disrupt host membranes is supported in a number of bacterial systems in which the loss of specific virulence functions abrogates NLR signaling. For example, in resting macrophages, cytosolic recognition of L. monocytogenes
requires the pore-forming toxin, listeriolysin O 
. Similarly, Helicobacter pylori 
and Legionella pneumophila 
mutants lacking a functional type IV secretion system (T4SS), and Shigella flexneri 
or Salmonella enterica serovar typhimurium 
mutants lacking a functional type III secretion system (T3SS) fail to stimulate NLR pathways. In each case, the virulence system in question is responsible for host membrane damage and the likely translocation of bacterial products into the cytosol where they can be recognized by NLRs and/or other cytosolic surveillance systems.
Similarly, we found that the ESX1 specialized protein secretion system of Mtb is required for Nod2 recognition. While it has been suggested that type I IFN induction via ESX1 might represent a specific immunomodulatory virulence strategy 
, analogies to these other pathogens suggests that perhaps NLR recognition is simply a byproduct of a membrane damaging function that allows bacterial products to enter the cytosol. This model is supported by our observations that other membrane perturbing agents, such as SLO and PANX1 can substitute for ESX1 function and allow cytosolic recognition. Thus, in a number of cases it appears that NLRs can be considered as sentinels for pathogens that rely on membrane damage as a pathogenic strategy.
Based on their common role in protein secretion and in facilitating cytosolic recognition, it is tempting to speculate that ESX1 and Gram-negative T3SS and T4SS function analogously to deliver effector proteins into the host cytosol. Despite these similarities, the role played by ESX1 during infection remains unclear, since no translocated effectors have been identified to date. In both Mtb and M. marinum
, a related pathogen of ectotherms, ESX1 has been implicated in host membrane disruption and one of the major substrates of this system, EsxA, has been proposed to possess a membrane-lytic activity 
. This single activity could be sufficient to account for the delivery of MDP and other PAMPs to the cytosol. It remains to be determined whether perturbing host membranes is the only role played by ESX1 during infection, or if this system also serves additional functions analogous to the specialized secretion systems of other pathogens.
A major consequence of the cytosolic recognition of Mtb is the induction of type I IFN. While the importance of this response in viral defense is clear and virtually universal, its role in antibacterial immunity appears to vary. Mice deficient in the type I IFN receptor, Ifnar1, are significantly more susceptible to several Gram-positive and -negative bacterial infections 
, indicating that IFNα/β are important for immunity to many bacteria. However, Ifnar1 mutation has the opposite effect on the outcome of L. monocytogenes
, suggesting that IFNα/β can also exacerbate disease. The role played by IFNα/β in Mtb infection remains somewhat uncertain. The induction of several immunologically important genes, including NOS2, depend on IFNα/β, suggesting a protective role. Initial studies of mouse and human infections appeared to support this view 
. However, like the L. monocytogenes
system, mutation of the IFNα/β receptor has in most cases been associated with decreased bacterial burden in mouse models of tuberculosis 
. IFNα/β may fail to protect against disease because Mtb inhibits the response to these cytokines in infected macrophages 
. The ultimate influence of IFNα/β on Mtb infection appears to depend on a number of experimental factors, which might include host species, bacterial strain, route of infection and dose. Despite these differences however, some important themes emerge from these studies. Most importantly, the effect of IFNα/β is most apparent after the onset of adaptive immunity and not before, suggesting that the major role-played by type I IFNs during tuberculosis may be to instruct the priming or maintenance of the adaptive immune response and perhaps to control the differentiation of regulatory T cells 
A variety of bacterial pathogens trigger the type I IFN response, and a paradigm has begun to emerge regarding the induction of this response by bacteria. One current model suggests that bacterial DNA translocated into the host cytosol is the major eliciting agent. This model is based largely on the observations that infection with L. monocytogenes
or L. pneumophilla
, or transfection of DNA into the cytosol induces a similar IFNβ response that is Rip2 independent, and Tbk1- and Irf3-dependent 
. Other PAMPs, such as MDP, can provide a synergistic IFN-inducing stimulus, but have not appeared to be sufficient for induction of IFNβ in the absence of other triggers 
In contrast, our data support a model whereby Nod2 stimulation by Mtb infection induces the polyubiquitination of Rip2, which acts via the Tbk1 kinase to stimulate the activity of Irf5 and induce transcription of IFNα/β. This differs from the pathway triggered by other bacteria such as L. monocytogenes
, which depends entirely on Irf3 in resting macrophages 
and does not involve Irf5 (). Although Irf5 has previously been shown to be activated by the MyD88-dependent TLR7 and TLR9 pathways, this work reveals a novel role for this protein in Nod2 signaling, and a new link between Nod proteins and the type I IFN response. Furthermore, we found that unlike the N
-acetylated MDP found in many bacteria, stimulation with the N
-glycolylated MDP derivative found in mycobacteria was sufficient to stimulate the IFN response in the absence of other stimuli.
A significant component of IFNβ induction remains intact upon Mtb infection of Rip2-deficient macrophages ( and ), indicating that additional pathways are also involved. Since virtually all IFNβ expression is ESX1-dependent, it appears that the residual induction observed in rip2−/−
macrophages also depends on cytosolic recognition pathways. These pathways could certainly include a DNA sensor that acts via Irf3, as proposed for other infections, since Irf3 deficiency had a moderate effect on IFNβ expression in our experiments ( and S2
). Thus, our data do not imply that Mtb is stimulating IFNα/β in a fundamentally different manner from other bacteria. Instead, it is likely that bacterial pathogens stimulate the IFN response via multiple, partially redundant pathways, and that the relative importance of each is determined by the unique biology of the infection. In the case of Mtb, we speculate that the N
-glycolylation of its peptidoglycan, and perhaps a paucity of other stimulants such as DNA, favor recognition via Nod2. It is also possible that the balance of these pathways might be affected by the activation state of the macrophage. When resting macrophages are infected with L. monocytogenes
, the IFN response requires LLO and is completely Irf3 dependent. In contrast, IFNγ-stimulated cells are able to deliver this bacterium to the lysosome, where the cell wall is degraded to produce abundant peptidoglycan fragments. In this situation, a significant component of the IFNβ induction depends on Nod2 and not Irf3 
. While Irf5 was not investigated in this study, it is possible that this represents another situation in which robust Nod2 signaling promotes a Nod2- and Irf5- dependent type I IFN response.
While we found that loss of Nod2-Rip2 signaling only partially reduces the induction of IFNβ, Rip2 deletion completely abrogated IFNα and RANTES expression. These results can be explained by the structure of the IFN regulatory circuit. Initially, only IFNβ is expressed, and subsequently IFNα and other interferon regulated genes (IRGs), such as RANTES, are induced via an Ifnar1 and Irf7-dependent autocrine/paracrine signaling pathway 
. Thus, it appears that the decrease in IFNβ expression that we observe is sufficient to severely impair downstream IRG induction, at least in this cell culture model.
Multiple steps of this pathway are likely to depend on stable ubiquitin modifications. Not only did we observe that Rip2 is polyubiquitinated upon infection, but we also found that a Rip2 point mutant that cannot be stably ubiquitin modified is unable to mediate IFNα/β induction in response to Mtb infection (Figure S5
). Collectively, these data suggest that polyubiquitinated Rip2 is required for Mtb-induced type I IFN expression via Irf5. Interestingly, MyD88-dependent activation of Irf5 involves formation of a tertiary complex that includes the E3 ubiquitin ligase, Traf6 
. This E3 ubiquitin ligase associates with Rip2 upon MDP stimulation, raising the possibility that a Rip2-Traf6-Irf5 complex might exist and that the activity of Irf5 might also be regulated by ubiquitin.
The specificity of the innate immune system has been shaped by the very powerful natural selection imposed by microbial pathogens. Our work suggests that upon infection with Mtb, a particularly potent form of MDP is translocated into the host cell cytosol where it triggers a novel signaling pathway leading to the robust induction of the type I IFN response. It is unlikely to be coincidental that the active component of our most potent adjuvant, complete Freund's adjuvant (CFA), consists of mycobacterial cell fragments. The specific pathway described in this work might play a major role in this adjuvant's effectiveness, since IFNα/β production is required for CFA to promote antigen-specific immune responses (55). Thus, while PAMPs are often regarded as invariant microbial components, it is clear that functionally important pathogen-specific differences exist in the composition of these molecules, and that the immune system can differentiate these subtly distinct structures.
Given the potent adjuvant activity of mycobacterial components, it is somewhat surprising that the attenuated vaccine strain M. bovis
BCG, which produces the same PAMPs present in CFA, provides poor protection against pulmonary TB in adults 
. The lack of ESX1 function in this strain appears to be at least partially responsible, since the reconstitution of ESX1 improves the efficacy of this vaccine 
. While this effect has previously been attributed to either the secretion of additional antigens or altered antigen presentation, it is also possible that ESX1 activity improves immunity by delivering crucial PAMPs into the cytosol where they are optimally recognized. Understanding both the details of PAMP trafficking, as well as the precise specificity of PAMP recognition, promises to aid in both the design of improved adjuvants and more effective tuberculosis vaccines.